Unreal Engine 5 (Product): news timeline & CVEs

Code2Games: Enabling Coding Agents for Gaming World Generation

Code2Games helps coding agents generate consistent game worlds and adapt them to Unreal Engine 5.

Code2Games is an agentic framework that turns a natural-language game intent into a structured world, starting from a Blender scene generated from the same intent. A shared scene-gameplay representation coordinates analysis, planning, constrained generation, and engine customization, then an execution-guided loop uses compilation, runtime, and gameplay-test feedback to adapt the world to Unreal Engine 5. On the ten-prompt GameCode4D benchmark it improves visual quality, interactive fidelity, artifact quality, and playable-game quality versus direct coding-agent generation.

Hugging Face daily papersOct 3, 20:00 UTC · 5d agoAI research

WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents

WorldAuditBench tests multimodal agents at finding 3D-world anomalies; frontier models reach 6.6–42.3% versus humans at 83.4%.

WorldAuditBench is a benchmark for auditing interactive 3D worlds for anomalies such as floating objects, traversable walls, and scene-inconsistent objects. It includes 213 tasks across 13 environments built with Unreal Engine 5 and Three.js, covering five anomaly families. Five frontier models were tested under a fixed exploration budget in two setups: VLA exploration followed by VLM identification, and an end-to-end VLM agent. Success rates ranged from 6.6% to 42.3%, well below human performance of 83.4%.

arXiv cs.AI / cs.LG / cs.CLSep 30, 17:55 UTC · 8d agoAI research1

WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents

WorldAuditBench shows multimodal agents trail humans when auditing anomalies in interactive 3D worlds.

WorldAuditBench is a benchmark for finding anomalies in interactive 3D worlds, with 213 tasks across 13 environments built in Unreal Engine 5 and Three.js and five anomaly families. Tasks require both navigation and visual reasoning, such as spotting floating objects, traversable walls, or scene-inconsistent objects. Five frontier models were tested under a fixed exploration budget in two setups: VLA exploration followed by VLM identification, and an end-to-end VLM agent. Success rates ranged from 6.6% to 42.3%, well below human performance of 83.4%.

Hugging Face daily papersSep 29, 20:00 UTC · 9d agoAI research

Unreal Engine 5

Timeline3 articles

Appears with

Code2Games: Enabling Coding Agents for Gaming World Generation

WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents

WorldAuditBench: Interactive 3D World Auditing with Multimodal Agents